towards sustainability of land use in a highly vulnerable and degraded

Bui Le Vinh Towards sustainability of land use in a highly vulnerable and degraded tropical soil landscape of northern Vietnam – bridging scales Auf dem Weg zu einer nachhaltigen Landnutzung in einer sehr anfälligen und degradierten tropischen Bodenlandschaft des nördlichen Vietnam – Skalen Überbrückung Hướng tới sự bền vững trong sử dụng đất cho vùng đất bị thoái hóa và dễ bị tổn thương ở vùng đồi núi phía bắc Việt Nam – Mô hình mở rộng This thesis was accepted as a doctoral dissertation on December 17th, 2014 in fulfilment of the requirements for the degree “Doktor der Agrarwissenschaften” by the Faculty of Agricultural Sciences at Hohenheim University. Date of oral examination: March 24th, 2015 Examination Committee: Head of the Committee: Prof. Dr. Jens Wünsche Supervisor and Review: Prof. Dr. Karl Stahr CoReviewer: Prof. Dr. Joachim Müller Additional examiner: PD Dr. rer. nat. Daniela Sauer

Towards sustainability of land use in a highly

vulnerable and degraded tropical soil landscape

of northern Vietnam - bridging scales

PhD dissertation, 2015, Bui Le Vinh

INSTITUTE OF SOIL SCIENCE AND

LAND EVALUATION UNIVERSITY OF HOHENHEIM

Soil Science and Petrography Unit

Prof Dr Karl Stahr

Towards sustainability of land use in a highly vulnerable and degraded tropical soil landscape of northern Vietnam – bridging scales

Dissertation Submitted in fulfillment of the requirements for the degree "Doktor der Agrarwissenschaften"

(Dr.sc.agr in Agricultural Sciences)

requirements for the degree “Doktor der Agrarwissenschaften” by the Faculty of Agricultural Sciences at Hohenheim University

Examination Committee:

1 Introduction 1

1.1 Problem setting 1

1.2 Objectives 2

1.3 Hypotheses 3

2 Methodology 5

2.1 SOTER approach 5

2.2 Geographic Information System 8

2.3 Data collection and derivation 9

2.3.1 Secondary data 9

2.3.2 Primary data 10

2.3.3 Derivation of some terrain variables using GIS techniques 14

2.3.3.1 Generation of main slope positions 14

2.3.3.2 Generation of curvatures 16

2.3.3.3 Generation of slope forms based on the main slope positions 17

2.3.4 Soil property calculation for Yen Chau 20

2.3.4.1 Calculation of further physical properties 20

2.3.4.2 Calculation of further chemical properties 21

2.3.4.3 Criteria for soil quality analysis 22

2.4 Soil mapping under fuzzy logic and Soil-Land Inference Model (SoLIM) 23

2.4.1 Limitations of conventional soil mapping under crisp logic and introduction of fuzzy logic soil mapping 23

2.4.2 Theoretical basis for soil inference using fuzzy logic 26

2.4.2.1 Basic theory 26

2.4.2.2 Expert system approach 27

2.4.2.3 Fuzzy set theory 28

2.4.3 Methodology 29

2.4.3.1 Knowledge acquisition 29

2.4.3.2 Soil-environment key development interview 30

2.4.3.3 Soil-environment description interview 31

2.4.3.4 Optimality curve definition interview 32

2.4.3.5 Knowledge verification interview 33

2.4.3.6 The fuzzy soil inference process 33

3 General description of the study area 35

3.1 Physiography 35

3.2 Geology 37

3.3 Climate 42

3.4 Soils and Land use 44

3.5 Ethnic groups and land use systems 46

4 Results 49

4.1 Characterization of the SOTER database of Yen Chau district 49

4.1.1 Terrain units 49

4.1.1.1 Characterization of major landforms and terrain units 49

4.1.1.2 Determination of terrain units for each of the parent materials 51

4.1.2 Terrain components 58

4.1.3 Soil components 60

4.2 Soils and soil properties 66

4.2.1 Overview of soils of Yen Chau 66

4.2.2 Major soil properties for soil quality assessment 74

4.2.2.1 Air capacity – AC (%) 75

4.2.2.2 Available water capacity – AWC (l/m2) 75

4.2.2.3 Organic matter – OM (kg/m2) 78

4.2.2.4 Total nitrogen – Nt (kg/m2) 78

4.2.2.5 Available phosphorous – PBray1 (g/m2) 79

4.2.2.6 S-value 80

4.2.2.7 The sum parameter N-P-S 80

4.2.2.8 Biological activity 82

4.2.2.9 Human impact 83

4.2.3 Computations of correlation coefficients for soil properties 84

4.2.4 Major soils and their distribution in Yen Chau 93

4.3 Soil mapping model 109

4.3.1 Calibration of the formation of soils in Yen Chau 109

4.3.2 Spatial delineation of the soil map of Yen Chau using SoLIM software 109

4.4 Soil quality mapping model 117

5 Discussion 127

5.1 The Yen Chau SOTER database 127

5.2 The variability of Yen Chau soils 128

5.2.1 Soil pH (H2O) 129

5.2.2 The A-horizon thickness 130

5.2.3 Soil organic matter 130

5.2.4 Total nitrogen - Nt 132

5.2.5 Base saturation 134

5.2.6 Cation exchange capacity for clay minerals (CECclay) 136

5.2.7 S-value 138

5.2.8 Standardized parameter N-P-K 139

5.2.9 Soil forming processes to the variability of soils in Yen Chau 140

5.2.9.1 Clay illuviation 140

5.2.9.2 Clay formation 142

5.2.9.3 Humus accumulation 143

5.2.9.4 Decalcification 144

5.2.9.5 Base leaching 145

5.2.10 Environmental conditions and their role in soil and soil quality mapping 147

5.2.10.1 Climate in association with elevation 147

5.2.10.2 Parent material 149

5.2.10.3 Relief 151

5.2.10.4 Vegetation 154

5.2.10.5 Biological activity 155

5.2.10.6 Human impact 155

5.2.10.7 Time 156

5.3 Validation of the soil and soil quality maps 158

5.3.1 Validation of the soil map 159

5.3.2 Validation of the soil quality map 159

6 Conclusions 163

6.1 General conclusions 163

6.2 Specific conclusions 163

7 Summaries 165

7.1 Summary 165

7.2 Zusammenfassung 168

7.3 Tóm tắt 171

8 References 175

8.1 Literature 175

8.2 Other information sources 188

9 Appendix 191

9.1 Abbreviations 191

9.2 Description of reference soil profiles 192

9.3 Soil properties calculations for the Yen Chau SOTER database 217

Acknowledgement 223

1 Introduction

Northwestern Vietnam is a mountainous region and home to almost three million people from many different ethnic minorities The region has remained the poorest over the whole country for many years with the highest poverty rate (Phan, 2008)

The region has a wide range of elevations, strong relief variations, land use patterns, climatic patterns, land cover, and petrography The geological patterns of the area are ophiolite complex, granitic intrusion complexes, volcanic rocks, terrigenous and carbonate sedimentary rocks ranging from Proterozoic age till today (Hung, 2010b)

Swiddening and slash-and-burn agriculture had a long cultivation history of people in mountainous regions and they had been claimed to be sustainable (Dao, 2000) These cultivation practices worked very well in providing efficient subsistence to local people and sustaining the land use systems of the area (Vien et al., 2004) However, due to population growth and immigration happening over decades, like many other mountainous regions, the population in the northwest of Vietnam has risen remarkably, creating a severe stress on cultivated soils in covering the food demand for the growingpopulation Moreover, upland agriculture has been shifted towards meeting markets’ demands, or market-orientation, i.e intensification of cash crops like maize as a leading income crop (Clemens et al., 2010) This has promoted more intensive uses of soils and deforestation to widen arable land on hill slopes Consequences have then quickly arrived represented by serious flooding due to increasing deforested area, increasing soil degradation due to soil erosion and soil nutrient depletion because of overuse of agro-chemicals and intensive use of agricultural land, and, therefore, decreasing productivity of the cultivated soils (Toan et al., 2001, Wezel et al 2002) The demand to mitigate these facts is to drive cultivation practices towards sustainability Some solutions to achieving the sustainable development goal are to recover the lost forested area and importantly to adapt new suitable crops that not only bring high income to the local people but also function to mitigate soil degradation and recover soil quality To do so, a good knowledge of soil resources and soil quality must be necessarily achieved (Igué, 2000)

1.1 Problem setting

A common database about different soils, their properties and distribution for the whole country has not been completed Soil information achieved so far is very sparse And not an exception, the northwest of Vietnam does not have a good soil database necessary for related research or applications (Bo et al., 2002; Tin, 2005)

a set of technical and economic requirements for land surveys, soil mapping and evaluation of soil quality For soil mapping, soil sampling in mountainous regions is carried out within a minimum areal unit of 30 ha with 5 soil profiles studied, about 1 profile for 5 ha on average Physical properties are described for all 5 profiles and only 1 of them is sampled for chemical analyses Therefore, there maximum 5 different soils can be classified for an area of 30 ha

With 6 soil profiles studied for one catena, Clemens et al (2010) found 4 soil types that likely have been formed by different distinctive combinations of slope position, slope inclination, elevation, and period of cultivation According to our soil survey experience, an average catena has an area of 3-10 ha Thus, with an area of 30 ha we can have at least three adjacent catenas which could end up with having more than 5 soil types found when applying the

potentially will produce a huge overgeneralization of soil information It means a lot of soil information in a huge area will be lost, if this method is applied Therefore, it is necessary to have another soil mapping approach that can better achieve soil knowledge for a mountainous region of Vietnam

Modern soil mapping methods have focused on studying relationships between soils and their forming factors (Odeh et al., 1991; Zhu et al., 1996, 1997a; Batjes et al., 1997, 2000, 2008; Schuler et al., 2008; Qin et al., 2009, 2012) Jenny (1941) mentioned five main soil-forming factors: parent material, climate, organisms, topography, and time In a study about the development of soils on hill slopes in a South Australian subcatchment related to the environment, Odeh et al (1991) emphasized the importance of soil-landform interactions in the local pedogenesis They concluded that landform features such as slope gradient, plan convexity, profile convexity contribute remarkably to spatial variations of soils The combinations of profile-plan curvatures result in different slope forms (surface forms) which will be discussed in details in section 2.3.3 Slope forms were taken into account in studying soil variations at large scales (subcatchment scale to regional scale) in some other studies (Schuler et al., 2008; Qin et al., 2009, 2012; Cong, 2011) Qin et al (2009) developed an

(2011) and Schuler et al (2008) also considered slope forms in mapping soils at subcatchment and catchment scales in mountainous areas having much stronger relief conditions However, they simply described some major slope forms, for example straight-straight, convex-convex, concave-concave, etc., that were observed in their soil observations, rather than trying to capture an overall image of slope forms for the whole areas Therefore, this information might not be sufficient in drawing a good picture of soil-landscape relationships Especially when soils of a larger area (ex 1:25.000 or smaller scale) having strong relief conditions are to be mapped from catena units

Terrrain Digital Database (SOTER) for the catchments as the basis for the implementation of his land evaluation work

The overall goal of this study over the last 4 years was to develop a detailed soil information system for the northwestern mountainous district of Yen Chau using SOTER database upscaled from Cong’s work (2011) From this upscaled SOTER database, a soil map and a soil quality map will be generated for Yen Chau district using Soil and Landscape Index Model (SoLIM) To achieve this goal, the following objectives have been identified:

of all possible slope forms that will be used later as one of the soil-forming environmental parameters in the SoLIM model in mapping the variations of different soils and the quality

of these soils

through unique combinations between soils and environmental conditions such as landforms, slope forms, slope gradients, parent materials

between soils and terrain characteristics from the SOTER database

parameters for the development of a soil quality map for Yen Chau applying SoLIM

1.3 Hypotheses

In the efforts to the mapping of soils and their quality in Yen Chau district, the following main hypotheses need to be proven

a) Different rock types have certain influence on soil occurrences and soil quality

b) Slope inclination affects the degree of soil loss due to erosion, therefore, the soil quality,

and leads to transformations to new soils in relation to human impacts through agricultural activities

c) Elevation plays an important role in the occurrences of different soils and their quality

degrees

d) Slope forms are very important in capturing spatial variations of soils and if they are

correctly spatially delineated, loss of soil information can be minimized and spatial soil gradation can be better seen as a continuum

e) The age of cultivation in relation to the major slope positions (crest, upper-, middle-, foot

slope, and valley) has influences on soil distribution and soil quality

2 Methodology

2.1 SOTER approach

Oldeman (1993) stated that degradation and pollution of land and water resources under pressure of increasing population had led to a need of having a system, which was capable of managing natural resources data so they could be accessed, combined and analyzed This management manner must serve potential use in the sense of meeting food requirements, mitigating environmental impacts, and maintaining environmental conservation From this fact, a Methodology for a World Soils and Terrain Digital Database (SOTER) was developed

at the International Soil Reference and Information Centre (ISRIC), based in Wageningen, the Netherlands, in 1987

The original idea and premise of SOTER were developed in Russia and Germany to map land characteristics based on interactions among physical, chemical, biological and social phenomena over time (van Engelen, 1995) SOTER studies land through distinctive combinations of soils and terrain characteristics such as landform, surface form, slope, and parent material Each combination represents a SOTER unit Data in the SOTER database are organized in:

different spatial representations of soil information

In addition, the SOTER database is incorporated with a set of rules, formulas, and models, which can be used to produce new maps, make scenario predictions, and derive new data SOTER allows developing databases on different mapping scales and it is possible for individual databases on different scales later to be merged into a global database (van Engelen, 1995) Beyond the existing map of establishing SOTER databases of the world, other studies have been carried out at different scales Oliviera and van den Berg (1992) first implemented a SOTER database in São Paulo State of Brazil at a scale of 1:100.000 Dobos et al (2005) carried out the development of a SOTER database at scales of 1:1 and 1:5 million for Europe Our working group at the Institute of Soil Science and Land Evaluation, Hohenheim University, Germany, has produced different SOTER databases at various scales Graef et al (1999) carried out a land evaluation study using SOTER at regional and village scale Gaiser

et al (1999) applied the SOTER approach for estimating yield potentials at regional scale in Brazil Graef and Stahr (2000) used the SOTER approach for management of soil, terrain, and land use and vegetation data at regional scale Herrmann et al (2001) applied SOTER as a tool for land use planning in West Africa at a scale of 1:200.000 Graef et al (2002) developed a SOTER database in Niger for improving soil and water conservation measures Igué (2000) and Igué et al (2004) successfully developed a SOTER database for Central Benin at a scale

of 1:50.000 Schuler et al (2010) and Cong (2011) created SOTER databases for small

the variability of physical and chemical soil characteristics in relation with landscape in Benin using the SOTER approach This research will develop a SOTER database for a mountainous district of northern Vietnam, at a scale of 1:25.000

SOTER units reveal unique combinations of soils and terrain characteristics and they can be mapped The information of these units can be stored in two ways: geometry and attribute data

in which an attribute characterizes an object or a geometric shape Geometry data are stored in

a Geographic Information System (GIS) Attribute data are structured progressively in the order of terrain units, terrain components and soil components, known as SOTER differentiating criteria (van Engelen, 1995; Oldeman and van Engelen, 1993) The structure of

a SOTER database is well illustrated in Figure 2.1, in which:

represents landforms of the earth’s surface When observing terrain characteristics, one should be able to capture as fully as possible the major landforms The major landforms can then be subdivided in combination with parent material or lithology Terrain units characterize an area through combinations of landforms and lithology A terrain unit can have one or more terrain components

parameters like surface forms, slope categories, mesorelief, surface drainage, ground water, etc A terrain component can have one or more soil components

profiles Every soil component has one or more fully described and analyzed reference soil profiles One soil profile should have maximum five subjacent horizons to the depth of at least 150 cm A soil horizon should not exceed 50cm in depth In the soil database, each horizon must be characterized with physical and chemical properties Reference profiles are represented on maps as points given by unique coordinates

2.2 Geographic Information System

Geographical Information Systems (GIS) is a computer-based information system, which allows representing, manipulating, storing, and analyzing features on the Earth’s surface, that are geographically and spatially referenced Since originated, it has been defined in many different ways by different authors and some of the definitions are as below with which a GIS is defined as:

 “a special case of information systems where the database consists of observations on spatially distributed features, activities, or events, which are definable in space as points, lines, or areas A GIS manipulates data about these points, lines, and areas to retrieve data for ad hoc queries and analyses” (Dueker, 1979) “a powerful set of tools for collecting, storing, retrieving at will, transforming and displaying spatial data from the real world” (Burrough, 1986)

 “a database system in which most of the data are spatially indexed, and upon which a set of procedures operated in order to answer queries about spatial entities in the database” (Smith et al., 1987)

transform geographically referenced spatial data into information on the locations, spatial interactions, and geographic relationship of the fixed and dynamic entities that occupy space in the natural and built environments” (Weller, 1993)

 “automated systems for the capture, storage, retrieval, analysis, and display of spatial data” (Clarke, 1997)

GIS has become a huge interest worldwide and has been applied in many different areas (Maguire et al., 1991) In agriculture, GIS was used in studying hot spots to deforestation (van Laake, 2004), planning manure application (Basnet et al., 2002), identifying potential irrigated agriculture in Western Desert of Egypt (Ismail et al., 2012), risk assessment of agricultural chemicals (Verro et al., 2002), modelling of soil contamination (Kumar and Vaani, 2008), or mapping soil nutrients (Yang and Zhang, 2008), etc

In military, GIS can be used in modeling of military operations such as movement of military personnel, detection of enemy (Pincevičius et al., 2005); studying the mobility of military vehicles on the ground through a relief analysis (Pahernik et al., 2006); tracking down a target

in an area where the target is likely to be located by using a fusion-based approach (Bardford

et al., 2011); etc

In health, GIS studies have been taken in public health (Xiaolin, 2006), studying contributing factors to malaria prevalence in Thailand (Jeefoo et al., 2009), modelling drinking water wells (Dursun et al., 2009), in health care analysis (Barnes and Peck, 1994), etc

In economic research, GIS has been used in selecting shopping mall location in Australia (Cheng et al., 2007), in mapping urban economic analysis (Clapp, 1997), using cost effectiveness analyses and cost benefit analyses to assess avalanche hazard mitigation

strategies (Fuchs et al., 2007), or in analyzing geo-demographic using a fuzzy clustering method (Son et al., 2012), etc GIS has a wide range of applications and can be used in many other different fields as well

According to van Engelen and Wen (1995), the soils and terrain database consists of information of soils and terrain characteristics in a Relational Database Management System (RDBMS) linked with a Geographic Information System (GIS) SOTER is where information

of terrain and soils is stored, and where each map unit representing a distinctive combination

of landform/terrain, parent materials and soil information is defined The GIS package is a tool

to spatially deal with questions that involve input data from the SOTER database (Burrough,

1986) Batjes (1990) said “the type of questions that can be asked is primarily determined by

the type, format and manner according to which attributes are stored in the database” To

provide answers to these questions, attribute and spatial data must be studied and analyzed thoroughly

The SOTER approach can be applied in many different aspects A detailed SOTER database can be used very well for implementations of land-use planning (Igué et al, 2004) and land evaluation (Cong, 2011) SOTER can be used in different modelling purposes, such as modelling of soil information (Schuler et al., 2010), hydrological processes and sediment yield (Bossa et al., 2012); calculating stocks of carbon and nitrogen in soils (Batjes et al., 1999; Batjes, 2000), mapping soil carbon stocks (Batjes, 2008), predicting soil organic carbon stocks and changes over a certain time (Batjes et al., 2007; Cerri et al., 2007)

2.3 Data collection and derivation

The geological map of Yen Chau district is part of the Vạn Yên geology and mineral

resources sheet (F-48-XXVII) which covers part of the northwestern Vietnam (Bao, 2004) The map has a scale of 1:200.000 According to the Van Yen sheet, the geology of Yen Chau originally has 18 units Due to many similarities in properties and characteristics as well as minor differences of the rocks, they were grouped into 8 different geological units

of VietNam, land-use planning at district and provincial level is implemented every 5

consecutive years The land-use map of Yen Chau was obtained on the completion for 2011

from the Department of Natural Resources and Environment of Son La province Most use maps in Vietnam were created from aerial photos and get updated periodically by the

land-network of cadastre offices and officers down to village level When updating a map, cadastre officers at lower levels gather information of changes of land uses and submit it to officers at higher levels, who are responsible for updating the changes at higher administrative levels (district, province)

2.3.2 Primary data

The maps mentioned in 2.3.1 and related information were very necessary for studying the area and making plans for field surveys based on altitude patterns, geological units, and major land use types Maps were studied and farmer meetings and interviews were carried out before

soil investigation was taken into implementation in order to have some overview about the

local soils through indigenous knowledge Soil surveys were then implemented very specifically for every single geological unit/rock type In each unit of geology, soils were studied at catena scale at five major slope positions (crest, upper slope, middle, foot slope, and valley) with an assumption that soils differ at different slope positions There were 5 to 7 soil profiles studied along a catena covering the 5 slope positions In total, there were 124 soil profiles described and analyzed for the Yen Chau SOTER database Furthermore, hundreds of auger drillings were taken as reference information to fulfill the soil database of Yen Chau The SOTER database was built by 3 PhD, 7 MSc, 3 BSc, and 4 internship students in the period of 2007-2012 within the SFB 564 Research Program

The soil profiles and auger drillings were described according to the Field Guide for Soil Description, Soil Classification and Soil Evaluation (Jahn et al., 2003) and the World

Reference Base for Soil Resources 2006 (FAO, 2006) In every soil horizon, physical properties such as soil colour, soil structure, soil texture, stone content, root density,

biological activity, etc were carefully described in the field Soil samples were then taken for laboratory analyses

Almost 600 soil horizon samples were taken The samples were taken to the lab in Yen Chau for being oven-dried (for Bulk Density measurement) and air-dried (for nutrient analyses) Air-dried samples were then ground, sieved to pass a 2 mm sieve and weighed Every sample bag weighed at least 400g and was packed so it remained dry when it got to the labs in Hanoi

and Hohenheim University for further chemical analyses The analyses were accomplished

available potassium

There were two automatic weather stations modeled CR 800 produced by Campbell

Scientific Ltd installed in Yen Chau district One station was installed in Muong Lum commune and the other one was set in Chieng Khoi commune The both weather stations recorded data of average air temperature, average relative humidity, average solar radiation, wind direction, wind velocity, and total rainfall A record was taken every two minutes

The achievement of physical and chemical properties (eg Table 2.1) is discussed as follows:

 Physical properties

Bulk Density (g/cm 3 ): five cylinders with each being of approximately 95 cm3 in volume were

for 24 hours and their dry weights were taken Bulk Density of these five core samples was calculated by dividing the dry soil weights by the volumetric cylinder The final Bulk Density value for the soil horizon was taken by the average number of the five values

Stone content (%): was estimated in the field using FAO guidelines for soil description and

soil classification (FAO, 2006)

Soil colour: soil colour was determined in the field under a moist condition using on a Munsell

colour chart (Oyama and Takehara, 1967)

Soil texture: soil texture was first carried out in the field under a suitably moist condition

followed with procedures described in Guidelines for Soil Description (FAO, 2006) Soil texture was then determined in a soil lab in Vietnam First soil organic matter was destructed

Soil structure: was determined in the field based on Guidelines for Soil Description (FAO,

2006)

Extraction of available P (P Bray1 _mg/kg): dissolve 11.11 g of reagent-grade ammonium fluoride

make to 10L volume with distilled water Mix thoroughly The pH of the resulting solution should be pH 2.6 ± 0.05 The adjustments to pH are made using HCL or ammonium hydroxide

Carbonate carbon (C carb _%): Ccarb was determined using a carbonate detector (Woesthoff Carmhograph C12S) A surplus of phosphoric acid was added to the sample in order to release

with 0.1M NaOH solution (Herrmann, 2005)

Exchangeable cations of Ca, Mg, K, and Na (mmol c+ /kg): Exchangeable cations were

flame photometer was used (Herrmann, 2005)

pH (H 2 O): was determined in a supernatant solution of a 1:2.5 soil-water mixture (FAO 1995)

The measurement was carried out with a WTW pH/mV Hand-Held Meter pH 330

pH (KCl): was determined in a supernatant solution of a 1:2.5 soil-1 M KCl mixture (FAO

1995) The measurement was carried out with a WTW pH/mV Hand-Held Meter pH 330

Potential cation exchange capacity (CEC_cmol c+ /kg): The soil was treated with Na-acetate in

order to exchange all cations Afterwards the sample was cleaned with ethanol To extract

propane activated flame photometer at 589m (Schlichting et al., 1995)

Total carbon (C t _%) and total nitrogen (N t _%): the content of Ct and Nt was determined with

a C/N analyser (LECO CN-2000) For samples without carbonate the soil organic matter

1995)

2.3.3 Derivation of some terrain variables using GIS techniques

2.3.3.1 Generation of main slope positions

There are 5 main slope positions: ridge, upper-, middle-, foot slope, and valley in which ridge and valley were extracted first as the basis for deriving the three slope positions in between

a) Gully (valley) position

A cell is considered a gully when its two opposite neighboring cells are higher in elevation and when its two orthogonal neighboring cells have one being lower and one being higher in elevation (Figure 2.2a) (Skidmore, 1990)

Figure 2.2 A stream line with raised channel sides and draining in a north-south direction (a)

Two cell wide stream line along and cross the drainage channel (b) (Skidmore, 1990)

The generation of a gully can be obtained by applying a function called Extract Drainage Networks in the SimDTA software written by Qin et al (2009) This software was provided personally by the author To operate this function, a corrected DEM file with 10m resolution

of the area was needed The function applies Peucker and Douglas (1975) algorithm

b) Ridge position

A cell is defined as a ridge when its two opposite neighboring cells are lower and its two orthogonal neighboring cells are either both being lower or have one being lower and one being higher in elevation (Figure 2.3a) (Skidmore, 1990) If the orthogonal neighboring cells have a lower cell and a higher cell in elevation, respectively, then the ridge is extracted with two cells equal in elevation (Figure 2.3b)

(b)

test cell

The generation of a ridge is obtained by applying a function called Extract Ridge in the SimDTA software To operate this function, a corrected DEM file with 10m resolution of the area is needed The function applies Peucker and Douglas (1975) algorithm

Figure 2.3 Ridge surrounded by lower cells (a) Ridge that drains in three directions which is

more than two cells long (b) (Skidmore, 1990)

c) Interpolating mid-slope positions

To create a map with every cell determined with a position relative to the ridge and to the valley of the slope containing that cell, Euclidean distance algorithm is applied To prepare for the interpolation of mid-slope positions (upper-, middle-, and foot slope), a valley and ridge must first be obtained The SimDTA software has a function called Relative Position Index which calculates the relative position of every cell to the ridge and valley of every slope This function applies the Relative Position Index (RPI) algorithm proposed by Skidmore (1990) as follows:

If a cell is not marked as a valley or ridge, the Euclidean distance from the cell to the nearest valley and the Euclidean distance from the cell to the nearest ridge are calculated These two distances are then summed up to calculate the ratio between the Euclidean distance of the cell

to the nearest valley and this summed distance to determine the relative position of the cell The calculation produces binary values within the range of [0, 1] in which 0 and 1 represented the lowest and highest positions along a catena, i.e valley and ridge, respectively From this range, smaller value ranges were divided to define the three mid-slope positions: upper-,

Orthogonal

to ridge Direction

of ridge

(b)

note that ridge drains

in three directions

ridgeline more than two cells long

middle-, and foot slope For the study area, the value assignment to characterize different slope positions is as follows:

2.3.3.2 Generation of curvatures

Curvature is defined as “the second derivative of a surface or the slope of the slope” (Muehrcke et al., 2009) In other words, curvature redraws the general shape of the land surface, or slope form, which is one of the very important soil formative factors (Odeh et al., 1992; Shary et al., 2002) Slope form is the description of the general shape of the slope in vertical and horizontal directions (FAO, 2006) There are two curvatures that characterize a slope form: profile (vertical) curvature and planform (horizontal) curvature

Profile curvature is parallel to the slope and indicates the direction of maximum slope and is

the rate of change of gradient It affects the acceleration and deceleration of flow across the surface and hence influences soil aggradation or soil loss (Odeh et al., 1991) (Figure 2.4a)

Planform curvature is defined as the rate of change of aspect being perpendicular to the

direction of the maximum slope and affects the convergence and divergence of flow across the surface (Odeh et al., 1991) (Figure 2.4b)

Figure 2.4 Profile (a) and Planform (b) curvatures (Aileen Buckley, Mapping Center

Lead-ESRI) (FAO, 2006) For both profile and planform curvatures, negative (-) values represent concave and positive (+) values represent convex forms; the value zero indicates a plane surface of that cell (Qin et al., 2009) Profile and planform curvatures can be computed by a function called curvatures in SimDTA software using Shary et al (2002) algorithm with a corrected DEM input file

2.3.3.3 Generation of slope forms based on the main slope positions

Figure 2.5 Nine major slope forms (redrawn from

Schoeneberger et al., 2002)

The Guidelines for Soil Decription

(FAO, 2006) mentioned nine major

Schoeneberger et al (2002) as can be

seen in Figure 2.5 A question is then

asked “How to characterize slope

forms of an area using these nine

Qin et al (2009) proposed a methodology in quantifying spatial gradation of slope positions

In this study, 11 slope positions were achieved and considered as a basic component of a landform The way to achieve this eleven slope position system is summarized in Figure 2.6 This two-tier hierarchical system of slope positions starts with the five major slope positions: Ridge, shoulder slope (upper slope), back slope (middle slope), foot slope, and valley The second tier is a subdivision of the first tier by taking into account the convexity and concavity

of the surface shape along a contour line

which the difference between the lowest point (233.6 m asl) and highest point (352.6 m asl)

However, the mountainous region of Yen Chau has extremely high relief conditions with elevation difference between the lowest and highest points of over 1400 m and highest slope

forms, which were observed during field surveys Hence, the 11 slope position system proposed by Qin et al (2009) would apparently not be appropriate for the situation of Yen Chau There must be another system that is more suitable in this case

Planar SHD (PSHD)

Convergent SHD (CSHD)

planar

Divergent BKS (DBKS)

Planar BKS (PBKS)

Convergent BKS (CBKS) concave

Divergent FTS (DFTS)

Planar FTS (PFTS)

Convergent FTS (CFTS)

Figure 2.6 System of slope positions (Qin et al., 2009)

We developed another second tier from the first tier of Qin’s system for Yen Chau Due to the high roughness of the land surface, all nine slope forms were depicted for each of the three mid-slope positions Therefore, the new second tier now has totally 27 slope forms for upper, middle, and foot slope positions, plus ridge and valley adding up to 29 slope forms To derive these 29 slope forms by a GIS package parameter value ranges were assigned for each of the slope forms as demonstrated in Table 2.2

There are three major forms: convex, planar or straight, and concave Therefore, value ranges were defined for these forms as follows:

These values were applied for both profile and horizontal curvatures The spatial delineation

of Yen Chau’s slope form system can be seen in Figure 4.3

2.3.4 Soil property calculation for Yen Chau

Reuler and Prins (1993) defined soil quality as the capacity of soil to provide plants with nutrients, water and oxygen, as well as rooting depth and rootability, heat regime, and stability

of the site These, altogether, guarantee the growth of plants Speaking of soil quality, it is a set of physical and chemical properties that are the supplies of nutrients, water, and oxygen in the soil, forming up the terminology of soil quality The primary physical and chemical properties were achieved during field surveys and after analyses in a soil laboratory as discussed in section 2.3.2

The nutrient stocks in the soil were calculated on the emphasis of effective rooting space (ERS) The ERS is the depth of the maximum reachable water by the roots during years of low rainfall, which has the average effective root depth limited to the first 30-40cm from the surface excluding O layers (FAO, 2006) In this study, we take the ERS of 100 cm

2.3.4.1 Calculation of further physical properties

Thickness is calculated for every soil horizon, which is the total depth of that soil horizon

Thickness is calculated by taking the large number on the vertical scale minus the small number as the below and upper boundaries of the horizon, respectively

Soil volume (l/m2) is a three dimensional measurement of a particular space for a certain amount of soil to fit in Soil volume is calculated as follows:

Soil nutrient stock (kg/m2) is the total weight of soil material in a volumetric unit and is calculated as:

Soil volume (l/m2) * Bulk Density (g/m3)*factor 1 (soil depth 0-30cm)

Estimation of some other physical properties such as Air Capacity (AC, in vol%), Available Water Capacity (AWC, in vol%), and Field Capacity (FC, in vol%) was achieved based on the values of texture, organic matter content and Bulk Density by looking up in a table of section 5.2 (page 171) in Field guide for soil description, soil classification and soil evaluation (FAO, 2006)

From this estimation, quantification of AC, AWC, and FC was calculated based on solid volume for every soil horizon by multiplying the estimated values with soil volume value divided by 100

2.3.4.2 Calculation of further chemical properties

CEC cmol c kg -1 of clay is the value used for classifying the four clay illuviated soils (Luvisols,

Lixisols, Alisols, and Acrisols) (WRB, 2006) and can be converted from the value of CEC

Humus (mg/kg) is calculated as: Corg (mg/kg) * 1.724, whereas Corg is organic carbon which is

The C:N ratio is the ratio of Corg to total nitrogen is and calculated by dividing Corg by Nt

Available nutrients, CEC, and exchangeable cations (mol/m 2 ): are calculated by multiplying

with Soil Mass/1000

Determination of the S-value: According to the Field guide for soil description, soil

classification and soil evaluation (Jahn et al., 2003), the portion of basic cations (Ca, Mg, K,

represents S-value The S-value (cmolc/kg) is calculated as: CECeff * BS/100 The estimation

horizons within the depth of 30cm is multiplied with a factor of 1, while the S-value for deeper horizons (>30cm) is multiplied with a factor of 0.5 because of its availability

2.3.4.3 Criteria for soil quality analysis

Air capacity (AC_%) and available water capacity (AWC_l/m2) are two important physical criteria which tell if the soil has enough water available for plants and reveal the infiltration rate of water deep into the soil (Jahn et al., 2003) AC tells us about the total air volume in the soil, partially the root penetration, and the process of organic matter decomposition AWC determines the amount of available water for plants that can be stored by the soil in the total root zone AC is calculated on the basis of soil horizons and the average value represents the figure of the ERS; whereas, AWC is calculated for the whole ERS by summing up the values

of different soil horizons

Nitrogen (N_kg/m 2 ): is the most important nutrient element for plant development, i.e

developing protein and the energy for the plants Moreover, nitrogen is the energy supply for soil microorganisms that work for the decomposition of raw materials into soil humus If there is

a deficiency of nitrogen in the soil, plant leaves turn yellow, grain yield and plant weight decreases (Ding et al., 2005) Therefore, knowing how much nitrogen is available in the soil, how much plants need, and how much should go to the decomposition process is very crucial

in predicting yields and recommending appropriate amount of N fertilizer necessarily added into the soil

development of energy, sugars, and nucleic acids P deficiency in the soil makes plants weak and their maturity delayed, resulting in low yields and low product quality (MacCauley et al., 2009)

S-value (mol/m2): is the portion of CEC occupied by basic cations (Ca, Mg, K, Na) in

formula:

The range-standardized sum parameter N-P-S, proposed by Mausbach and Seybold (1998), is

the minimum value to 0 of each of the three parameters for the whole set of soil profiles and

standardized values of N, P, and S-value

2.4 Soil mapping under fuzzy logic and Soil-Land Inference Model (SoLIM)

2.4.1 Limitations of conventional soil mapping under crisp logic and introduction of fuzzy logic soil mapping

Soils have been studied for a long time for their importance in agriculture In order to have correct and sufficient information of soils is crucial in succeeding any work that depends on soil Therefore, people started trying to make soil maps a long time ago Soil mapping approaches have had a history of development of over 100 years Early methods were focused

on genetic principles such as zonality, and emphasized the importance of knowing the continuous nature of soils (McBratney and De Gruijter, 1992) These were the milestone for

Figure 2.7 Discretization of soils in the parameter domain (a) Dots represent the

locations of soils in the parameter domain, rectangles represent the boundaries of soil

classes in the parameter domain (b) Dots represent the centers of soil classes, the

intervals between the projected centers on their respective axes represent the attribute

resolution on these property axes (Zhu, 1997)

modern soil mapping methods coming to fulfill the knowledge of soil as a continuum According to Zhu (1997a), the mapping process consists of two sub-processes: soil classification and soil mapping Soil classification sub-process is the stage where soil scientists study soils in the area to be mapped by conducting soil surveys, implementing field observations, describing and classifying soils, and studying environmental conditions that might have constituted to the formation of the soils that have been prescribed for the area In other words, soil classification is a process in which soil groups and their properties in the soil property domain need to be identified

The rectangles in Figure 2.7 represent areal property domains of three soil groups A, B, and C

properties falling within these property areas are assigned with soil names A, B, and C However, in reality soils found within these areas often have their properties deviated from the central property values of the defined soil groups (Figure 2.7a) And there can be soils that belong to different soil groups, but have very close property values, or soils in a same group but have very different properties The fact that soil mapping using crisp logic approach can assign only one value to one soil group (Zhu, 1997a) cannot portrait soils as a continuum (McBratney and De Gruijter, 1992) and, therefore, leads to great losses of soil information Soil mapping is a process in which areas are delineated and assigned with values representing soil classes (soil mapping units) (Zhu, 1997a) Zhu emphasized two types of generalization during the mapping process: class assignment generalization and spatial generalization

Class assignment generalization is a process in which soils that have similar property values

to those of a prescribed soil mapping unit are assigned to that unit Conventional soil maps are created using crisp logic method under which a soil at a given location can belong to one and only one soil group and can only have properties of that soil group This means two neighboring soils belonging to two different groups are said to be different in soil properties,

or to have same soil properties if they lay within the area of a same soil group This may provide wrong or insufficient information as some soil objects really have subtle differences in properties but they belong to two different soil groups

For example, the soil dots 1 and 2 in Figure 2.7a belong to soil groups B and A, respectively (Figure 2.7b) However, the locations of these two dots on the property scales show that they may be similar soils rather than being distinguished as two completely different soils Or within soil group A (Figure 2.7a), the soil dots 2 and 3 are quite far from each other and they don’t seem to have very similar properties However, under crisp logic, they are classified into

These pieces of information are then deviated from reality, leading to false soil information achieved and incorrectly spatially delineated later on the map

Spatial generalization is a process of delineating soil polygon areas geographically and

spatially onto a map This process involves map scales and mapping techniques The detail of

soil information in reality brought to appear on a soil map depends on a map scale or a minimum size on the ground which soil information can be presented on the map This minimum area in mapping is called spatial resolution and characterized by a square or cell (ex

10 m by 10 m) Areas on the ground larger than this size can be well presented on the map The spatial resolution of 10 m by 10 m is rather small and used for making maps at large scales only, meaning mapping small areas At smaller scales, much larger area to be mapped, the spatial resolution increases (ex 30 m by 30 m or higher), making a conventional soil map not able to describe information of areas smaller than this cell size Soil information of these small areas is then ignored and included into more dominant soil groups Therefore, a large percentage of soil information can be lost when producing soil maps at small scales

For instance, Figure 2.8 shows two soil maps at a large scale (left) and a small scale (right) At the large scale map, the spatial resolution is small; therefore, soil units C, D, E, and F are well delineated on the map with obvious differences to neighboring soil units A and B At the small scale soil map, the spatial resolution now is higher because the mapping area is larger The cell size now far exceeds the size of soil unit polygons C, D, E, and F, and they get included into soil units A and B This means that there is a loss of soil information, when mapping large areas or making soil maps at small scales

Figure 2.8 Representation of soil bodies at different scales (adopted from Zhu, 1997)

Class assignment generalization and spatial generalization are the two major limitations of conventional soil mapping approaches which create errors of soil information (attribute and spatial data) These approaches under crisp logic are not able to describe the continuity of soils

in reality, but rather interpret soils as sharp and abrupt polygons However, soils in reality often have gradual spatial variations and rather diffused than sharp boundaries (Mark and Csillag, 1989; McBratney, 1992) Therefore, conventional soil maps under crisp logic are not suitable

At a larger scale map

At a smaller scale map

for mapping soils as a continuum Alternative mapping techniques must be developed in meeting the description of soil continuity in which fuzzy logic has risen up as a suitable approach

This methodology is a combination of fuzzy logic, GIS and expert system development techniques to model the spatial variation of soil information based on constituting environmental conditions (Zhu et al., 1996) Every point in the mapping area is a so-called soil similarity vector (SSV) which denotes how similar it is to the point of a prescribed soil class Hence, soil property values calculated for different points can be intermediate to the central property values representing prescribed soil classes By applying this soil mapping method, losses of soil information can be minimized and a gradual representation of a soil continuum can be achieved

2.4.2 Theoretical basis for soil inference using fuzzy logic

2.4.2.1 Basic theory

Soil is originated from rock and the transformation of rock into soil is called soil formation During weathering processes, the internal space of the earth, where rock occupies, is at equilibrium state The outer space, where rock is in contact with the atmosphere, is normally unstable and impacted by external factors such as heat, rain, wind, root penetration, etc., and

weathering starts from here Over time, rock breaks down making the inner space exposed and weathering feasible deep down from the surface With the time factor, weathering processes

go on strongly at certain depth and deeper from the surface and by this time soil has been formed (Jenny, 1941)

Therefore, we can say that with the factor time soil is a final product of weathered rock in

which interactions of climate, parent material, organism, and topography take place The soil formation process is expressed in the following formula:

S = f(Cl, Pm, Og, Tp,M,t) (1)

where Cl is climate, Pm is parent material, Og is organism, Tp is topography, M is Man, and t is

time

Formula (1) is a set of factors that constitute to the formation of soil in general The formation

of different soils generally follows this function, but how these environmental conditions influence soil formation is very much complicated, when taking into consideration of specific conditions of geography, climate, parent materials, and topography It is extremely difficult to fully understand the relationship between soils and their forming factors for the complexity of the soil forming processes (Zhu et al., 1996) However, many soil scientists have successfully studied the relationship between soils and their forming factors and gained lots of precious knowledge and experience about this matter (Odeh et al., 1991, 1992; Zhu et al., 1996; Zhu, 1997a, 2000; Batjes et al., 1997, 2000, 2008; Bo et al., 2002; Tin, 2005; Schuler et al., 2008; Qin et al., 2009, 2012; Clemens et al., 2010;) These inherited knowledge and experiences together with the knowledge learned from field surveys are very crucial and precious for soil

mapping using SoLIM These are called experts’ knowledge, which is very important to define correctly the relationships between different soils and the formative environmental conditions

in the area

2.4.2.2 Expert system approach

Expert system is one of the six branches of artificial intelligence (Badiru and Cheung, 2002), which can be defined as computer-based software systems that use both facts and heuristics to

be capable of solving difficult problems based on some knowledge-rich domain (Hall and Kandel, 1992; Badiru and Cheung, 2002)

There are at least three parts typically making up an expert system: a working memory, a

knowledge base, and an inference engine (Kandel, 1991; Badiru and Cheung, 2002) The

working memory is used to store information achieved from the user of the system The

information in our research should be related to the study area such as geological units, parent materials, elevation, slope, major landforms, slope forms, and land cover, etc The working

memory is the most dynamic component in an expert system as its content or data structure

changes with specific problem settings (Badiru and Cheung, 2002)

The knowledge base is where the expert domain knowledge is needed to solve difficult decision making problems It has moderately dynamic degree because it does not need change,

only if some information comes up for a change in the problem solution procedure (Badiru and Cheung, 2002) It contains declarative knowledge about facts, events, phenomena and relationships among them (Zhu et al., 1996; Berge and van Hezewijk, 1999) For instance, the knowledge base in this study is the relationship between environmental parameters and the formation of different soils that have been prescribed in the area Declarative knowledge differs from procedural knowledge by the fact that it contains simple facts and events stored in

a knowledge base

The procedural knowledge, in fact being the inference engine, utilizes the declarative

knowledge to solve problems or achieve purposes (Siler and Buckley, 2005) In Kandel’s

words, “the inference engine uses the knowledge in a particular representation to come to

some expert conclusion or offer expert advice It contains the system’s general solving knowledge It is responsible for determining what piece of knowledge to use next and scheduling other necessary actions It will take all actions indicate, as necessary, by a piece of knowledge which is found to be true, due to the current facts represented to the expert system”(Kandel, 1991) The inference engine is the least dynamic component in an expert

problem-system because of its strict control and coding structure If necessary, changes are only made

to “correct a bug or enhance the inferential process” (Badiru and Cheung, 2002)

From the discussions above it is obvious that a successful operation of an expert system very much relies on the quality of the knowledge domain The question is then asked how to obtain the knowledge domain The process of acquiring domain-specific knowledge is called

knowledge acquisition and considered the key successfully operating an expert system (Gaines

and Shaw, 1991; Kandel, 1991) To complete this task, the knowledge engineer and the expert system developer normally must work together on extracting the expert’s domain knowledge that will be used later in the expert system (Kandel, 1991) There have been many studies on developing knowledge acquisition techniques (Zhu et al., 1996; Zhu, 1997a, 1997b, 199b, 2000), and application of the succeeded Soil Landscape Inference Model-SoLIM (Zhu et al., 1996; Zhu, 1997a, 199b; Schuler, 2010) will be used in this study

2.4.2.3 Fuzzy set theory

The term “fuzzy” is used to characterize classes that cannot or do not have sharply defined boundaries (Burrough, 1989) It is different from conventional or crisp logic sets which only allow binary membership functions (i.e true/yes (1) or false/no (0)), hence defining sharp boundaries of soil groups Fuzzy logic sets, on the other hand, allow the possibility to have gradual representation of classes denoted by membership values (Zhu, 1996), therefore, the boundaries of the classes diffuse As discussed in section 2.4.1., soil is a spatial continuum in which different soils rather have gradual changes in properties than sharp Assigning crisp values to different soils will lead to losses or insufficiency of soil information Therefore, fuzzy logic must be used to capture the spatial distribution of soils

A fuzzy set is defined (Zimmermann, 1976) as: If X = {x} is a collection of objects denoted generically by x then a fuzzy set A in X is a set of ordered pairs:

A = {x, A(x)} x  X (2)

membership of x in A which maps X to the membership space M (when M contains only the two

set) The range of the membership function is a subset of the nonnegative real numbers whose supremum is finite

There are many fuzzy set operations that can be used in a fuzzy set Within the scale of this study, we discuss only the following operations:

Intersection: The membership function N (x) of the intersection (logic ‘and’) set of fuzzy sets A

Union: The membership function N (x) of the union (logic ‘or’) set of fuzzy sets A and O is

Complement: The membership function N (x) of the complement (logic ‘not’) set of fuzzy set A

The equations (3) and (4) are the fuzzy minimum and maximum operators, respectively, and are the two main operators in the fuzzy inference process of fuzzy soil mapping, which will be discussed about, when they are used in more details in section 2.4.3.6

2.4.3 Methodology

As discussed in sections 2.4.1 and 2.4.2, soil is a spatial continuum and soils at different points

do not necessarily have identical property values to those of the central soil concepts, i.e values that defined the prescribed soil classes In other words, they have intermediate values in between these central values A soil at a location has properties similar to those of one or more soil classes of an area containing that location The similarity of soil at a specific location to one or more prescribed soil classes is called soil similarity vector (SSV) Mapping soils for an area using fuzzy logic method means deriving SSV for this area (Zhu 1996) This means this soil is likely to be named after a prescribed soil class to which it is the most similar, i.e having the highest similarity value (expressed in percentage) to that (100%) of this prescribed soil

is a pixel within a grid covering the area and having column number c and row number r

an unique combination of environmental conditions This combination is called an instance of

that soil class A soil class can have more than one instance For example, Luvisols in the area

may occur on volcanic rock at gentle to moderate slopes; or in limestone area at gentle slopes having straight straight landforms; or on sedimentary rock at gentle slopes at upper slope positions having straight straight landforms In this case, Luvisols have three instances

Environmental conditions of an instance of soil class k can be characterized as

E i k = (e i1 k … e il k … e im k ) (6)

k Inference of s k then has two sub-problems (Zhu, 1996): defining instances of the soil class

in the parameter space, and determining the similarity of a soil to the soil class at a point away from the typical occurrences of the prescribed soil class In other words, the second problem aims to figuring out the behavior of the soil class when the constituting environment variables deviate from the central defined values

conditions likely constituting to the formation of that soil class, when conducting a soil survey and/or knowledge from soil scientists who understand well about the relationship between the soil and the formative environmental conditions This knowledge is referred to as “Type 1” Determining the behavior of a soil class in response to changes of the formative environmental conditions is a more difficult task and is referred to as knowledge “Type 2” These two knowledge types can be obtained by a knowledge acquisition technique

2.4.3.1 Knowledge acquisition

Hoffman (1990) discussed and made a conclusion on three broad categories of knowledge

acquisition methods: task analysis, special task, and interview methods In task analysis

methods, knowledge engineers study tasks that are normally performed by experts In specific

task methods, tasks are categorized into different degrees of detail for specific analyses The

details of these tasks are very much dependent on the purpose of the analysis The interview

methods include 2 types of methods: unstructured and structured Unstructured interviews are

in fact free-flowing talks with open questions asked to collect knowledge from the experts Open questions mean letting the interviewees (experts) express their opinion rather than making an YES or NO confirmation about a specific topic, and they normally start with Wh, How questions Structured interviews require careful planning of questions and the order of their appearances, and they can reveal very much of experts’ knowledge Scientists in expert systems very often prefer using structured interviews to obtain the knowledge of experts (Zhu

et al, 1996)

Zhu divided the knowledge acquisition using structured interviews into four stages: the environment key development interview, the soil-environment description interview, the optimality curve definition interview, and knowledge verification interview In this study, we employed Zhu’s organizational division of the four structured interviews in obtaining the experts’ knowledge for inferring soil spatial variations for the study area

soil-2.4.3.2 Soil-environment key development interview

In key development interview, some key that can differentiate the soil classes based on their constituting environmental variables is highlighted This interview in fact is a step for the next interview, description interview An example of the structure of the key development interview in this study is as follows:

2.4.3.3 Soil-environment description interview

The interview is designed to extract the expert knowledge Type 1 from the soil scientists This

means it is used to describe the instances of a soil class, i.e occurrences under which typical environmental conditions likely constituting to the formation of that soil class are defined The description of environmental variables of different soil classes can be very confusing as two soils are similarly different maybe due to a small difference of a single (or more) environmental variable Thus, the knowledge engineer must carefully check the consistency of the information provided by the soil experts

Table 2.3 Description of typical instances of some soil groups

Slope

FT

FT-SS

FT

SH-VV, BK-VV, FT-VV

FT

SH-SS, BK-SS, FT-SS

FT

SH-VV, BK-VV, FT-VV

FT

SH-VV, BK-VV, FT-VV

SH, BK,

FT

SH-SS, BK-SS, FT-SS,

SH-VV, BK-VV, FT-VV

FT

SH-SS, BK-SS, FT-SS,

SH, BK,

FT

SH-SS, BK-SS, FT-SS,

SH-VV, BK-VV, FT-VV

FT

SH-SS, BK-SS, FT-SS

* Slope form (in vertical and horizontal order): SS – straight straight, VV – convex convex

After obtaining knowledge via description interview as shown in Table 2.3, the expert engineer must cross check this information with the information achieved from the key development interview If there is any inappropriate piece of information between the description and key data, the expert engineer may ask the soil experts to make this relationship right and reasonable

2.4.3.4 Optimality curve definition interview

The optimality curve definition interview is designed to extract knowledge Type 2 about the

behavior of a soil class in response to changes of its constituting environmental variables from its optimal instances (Zhu et al., 1996; Zhu, 1999, Zhu et al., 2003) The similarity value

ranges from 0 to 1.0 as the optimal value As discussed above, the knowledge Type 2 is

extracted for one variable at a time There are three basic forms of membership functions: shaped, S-shaped, and reverse S-shaped

bell-The application of the bell-shaped function to a variable means that the similarity value decreases from its optimal value when the environmental variable changes When the S-shaped function is applied, similarity value decreases when the value of the environmental variable is below the value where the optimal similarity reaches The application of the reverse S-shaped function means that similarity value decreases when the value of the environmental variable is higher than the value where the optimal similarity is obtained The illustration of these three basic functions is shown in Figure 2.9

Figure 2.9 The three basic forms of membership functions (Zhu et al., 2003)

The user can manually customize the optimality of an environmental variable by changing some critical points to fit the curve to the conceptual view of the soil experts Figure 2.10 illustrates an example of how different pieces of information can manually be put in the optimality curve In a particular case of Luvisols (A), the optimal value for having Luvisols is

Type 1 knowledge

Type 2 knowledge

Type 1 knowledge

Type 2 knowledge

mostly likely to be Regosols Soils on gentler or higher slopes are less likely to be Regosols

Figure 2.10 Optimality curves for slope gradient of instances 1&2 of Luvisols (A), instance 1

of Regosols (B), and instance 4 of Alisols (C)

2.4.3.5 Knowledge verification interview

When the knowledge Type 1 and Type 2 have been obtained, the knowledge verification interview is then applied to refine the knowledge set (Zhu et al., 1996) The knowledge verification process is carried out in two parts: indoor and outdoor For indoor verification, the expert engineer and the soil scientist work closely to check the result of inferred soil similarity maps with the knowledge of the scientist Changes are made if necessary to fit with the perceptual concept of the scientist about the soil map of the area until the results are satisfing During outdoor verification, the soil scientist goes back to the field to check on some confusing and difficult areas marked on the inferred soil map The scientist then carries out some more soil observations and analyzes the samples and incorporates this new data into the knowledge base to go on with the inference process The process will end when there is a reasonable result satisfactorily produced for these confusing areas

2.4.3.6 The fuzzy soil inference process

The inference process is illustrated in Figure 2.11 The inference engine is operated using a raster data approach in which fuzzy similarity values are calculated for every grid cell Zhu et

al (1996) assumed that the formation of a soil class at a cell is controlled by the least optimal environmental variable for the soil class at that cell Thus, Eq 3 fuzzy intersection (or fuzzy minimum operator) is used to calculate the integration of these influences from the environmental variables The inference engine takes a set of environmental variables derived from a GIS package for calculating the optimality curves

Then the minimum operator takes these optimal values for calculating the membership value for one instance of that cell In case the soil class has more than one instance, the set of

Slope Gradient (degree)

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1